Defining and validating predictors of cerebral infarction using brain CT images

Summary This project will contribute to a larger project on developing decision assistance tools for management of acute stroke. You will find and validate expressions of known clinical variables on CT images using imaging informatics, with Computer Vision and Deep Learning tools.

Project Details

Computed tomographic perfusion imaging (CTP) is a method of extracting the hemodynamic properties from the brain of a patient with suspect ischemia. These properties are used to determine the extent of irreversible damage (the ischemic core) and the extent of reversible damage (penumbra), that is, regions where restored blood flow will salvage the tissue. Treatment decisions can be made based on that information. However, the standard non-contrast CT (NCCT) also contains useful information which, combined with CTP measures will provide a more comprehensive analysis of tissue risk and, ultimately, a better standard of care.

In this project, you will process NCCT images and determine their predictive value using a large repository of retrospective patient imaging data. This will involve using computer vision tools such as skull stripping, image alignment and registration to atlas, using and testing quantitative measures of clinically known imaging characteristics such as hypodensity as predictive measures, as well as using Convolutional Neural Networks to find hidden features of the NCCT.

This is a suitable project for a student wishing to gain more experience in computational methods while working as part of a larger stroke research team.